Verification of maximum and minimum temperature forecasts from various agencies for Duluth, MN

The increasing availability of computing power and forecasting tools facilitates further progress in forecasting. In particular, different public and private entities disseminate deterministic maximum and minimum temperature forecasts through seven days out, with some agencies presenting fifteen days of lead time. It is more difficult, however, to find routine verification of such forecasts – to determine whether these long-lead forecasts have accuracy with respect to an accepted truth value and skill over climatology. There are numerous statistical methods to tease out the utility of given forecasts, each with its own advantages and weaknesses. Nevertheless, no single verification diagnostic seems to completely assess forecast quality; thus, numerous measures must be intercompared.

Accordingly, in this case study, forecast temperature maxima and minima were recorded for Duluth, MN from 1 February 2009 through 13 June 2009. Data were manually collected from the National Weather Service (NWS), the NCEP Global Forecast System (GFS) Model Output Statistics (MOS), The Weather Channel, Intellicast, Accuweather, and MyForecast publicly accessible websites. To measure forecast quality, various statistical techniques are employed to compare forecast outlets with each other, climatology, and verification. Not only are there expectedly diminishing returns with longer range forecasts, but there appears to be a general unwillingness or inability to stray from climatology. As such, the most utility may be derived from short term forecasts. The various synoptic patterns that are associated with forecasts successes and busts will be summarized.